Abstract: Objective To integrate parameters from conventional risk scores and apply machine learning （ML） algorithms to develop and validate a novel risk prediction model for in-hospital major adverse cardiovascular events （MACE） in patients with acute myocardial infarction （AMI），with the aim of providing a more accurate assessment tool for clinical risk stratification． Methods This study retrospectively enrolled 883 AMI patients who presented to the Chest Pain Center of the First Hospital of Lanzhou University between January and December 2019 and underwent emergent percutaneous coronary intervention （PCI）． Parameters required for the Global Registry of Acute Coronary Events （GRACE） score，Thrombolysis in Myocardial Infarction （TIMI） risk score，and Age-Creatinine-Ejection Fraction （ACEF） score were collected，with in-hospital MACE as the study endpoint． Three ML algorithms—Random Forest （RF），Light Gradient Boosting Machine （LightGBM），and Extreme Gradient Boosting （XGBoost）—were used to develop prediction models． Feature selection was performed using the Boruta algorithm，and model interpretability was assessed using SHapley Additive exPlanations （SHAP）． Model performance was evaluated using the area under the receiver operating characteristic curve （AUROC） and the area under the precision-recall curve （AUPRC）． Calibration curves were plotted to assess model calibration． Results Among the 883 AMI patients included，the in-hospital MACE incidence was 3.06％． Boruta feature selection identified Killip class，Left Ventricular Ejection Fraction （LVEF），serum creatinine （Cr），systolic blood pressure （SBP），cardiac arrest （CA），and heart rate （HR） as key predictors． In the testing cohort，the LightGBM model showed the highest discriminative ability （AUROC 0.93，95％ CI： 0.80－1.00），while the RF model achieved the best precision for MACE prediction （AUPRC 0.68，95％ CI： 0.30 － 0.94 ） with the greatest stability． Feature importance analysis identified Killip class as the most influential predictor，followed by Cr and LVEF． Calibration curves indicated accurate predictions in the low-risk range，with overestimation and underestimation in intermediate- and high-risk ranges，respectively． Conclusions ML models based on GRACE，TIMI，and ACEF parameters outperformed conventional risk scores in predicting in-hospital MACE among AMI patients． LightGBM exhibited the highest discriminative ability，RF was superior in identifying high-risk patients，and Killip classification，LVEF，and Cr were identified as core predictive features．